Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Forecasting patient length of stay in an emergency department by artificial neural networks

Forecasting patient length of stay in an emergency department by artificial neural networks 10.7603/s40690-015-0015-7 JOURNAL OF AERONAUTICS AND SPACE TECHNOLOGIES JULY 2015 VOLUME 8 NUMBER 2 (43-48) FORECASTING PATIENT LENGTH OF STAY IN AN EMERGENCY DEPARTMENT BY ARTIFICIAL NEURAL NETWORKS Muhammet GUL Ali Fuat GUNERI Yildiz Technical University, Mechanical Yildiz Technical University, Mechanical Engineering Faculty, Department of Industrial Engineering Faculty, Department of Industrial Engineering, Istanbul, TURKEY Engineering, Istanbul, TURKEY mgul@yildiz.edu.tr guneri@yildiz.edu.tr th rd Received: 18 April 2015, Accepted: 23 July 2015 ABSTRACT Emergency departments (EDs) have faced with high patient demand during peak hours in comparison to the other departments of hospitals because of their complexity and uncertainty. Therefore prolonged waiting times in EDs have caused the dissatisfaction on patients. Patient length of stay (LOS), also known as patient throughput time, is generally considered to be the length of time that passes from the patient’s time of arrival at the ED until time of discharge or transfer to another department of the hospital. Starting from patient admissions to the EDs it becomes important have to be known the overall LOS in terms of right resource allocation and efficient utilization of the department. For this purpose this paper aims to forecast patient LOS using Artificial Neural Network (ANN) within the input factors that are predictive such as patient age, sex, mode of arrival, treatment unit, medical tests and inspection in the ED. The method can be used to provide insights to ED medical staff (doctors, nurses etc.) determining patient LOS. Keywords: Forecasting, Patient Length of Stay, Emergency Department, Artificial Neural Networks. YAPAY Sø Nø R Aö LARI KULLANILARAK ACø L SERVø S HASTA KALIù SÜRESø Nø N TAHMø Nø ÖZET Acil servisler hastanelerin di÷ er birimlerine göre karmaú Õ klÕ k ve belirsizli÷ in fazla oldu÷ u birimler olduklarÕ ndan, yo÷ un saatlerde yüksek hasta talebi ile karú Õ laú maktadÕ rlar. Bundan dolayÕ , acil servis içerisinde uzayan bekleme süreleri hastalar üzerinde memnuniyetsizli÷ e yol açmaktadÕ r. Hasta kalÕ ú uzunlu÷ u, di÷ er bir deyiú le hastanÕ n geçirdi÷ i toplam süre, genellikle hastanÕ n acil servise geliú inden taburcu edilmesi ya da hastanenin di÷ er bir birimine sevk edilmesine kadar geçen sürenin uzunlu÷ u olarak de÷ erlendirilmektedir. HastanÕ n acil servise kabulünden baú lanÕ larak, toplam kalÕ ú uzunlu÷ unun bilinmesi do÷ ru kaynak tahsisi ve birimin etkin kullanÕ mÕ açÕ sÕ ndan önemli hale gelmektedir. Bu amaçla bu çalÕ ú ma, hasta yaú Õ , cinsiyet, varÕ ú türü, muayene ünitesi, acil serviste uygulanan tÕ bbi testler ve muayene gibi belirleyici olan girdilerle birlikte Yapay Sinir A÷ larÕ (YSA) kullanÕ larak hasta kalÕ ú uzunlu÷ unu tahmin etmeyi amaçlamaktadÕ r. Metot, acil servis tÕ bbi personeline (doktorlar, hemú ireler vb.) hasta kalÕ ú uzunlu÷ unun belirlenmesi için fikir vermede kullanÕ labilmektedir. Anahtar Kelimeler: Tahmin, Hasta KalÕ ú Süresi, Acil Servis, Yapay Sinir A÷ larÕ . urgent one is essential. But people don't have any 1. INTRODUCTION urgent cure may apply to emergency departments Emergency Departments (EDs) are the busiest rarely. These applications cause patient overcrowding departments in a hospital and its main purpose is to in many emergency departments. Thus waiting times provide timely emergency care to patients [1,2]. The and unsatisfaction levels of patients may increase, a highest attribute of emergency departments is their general complex condition may occur in the ED [3]. uninterrupted serviceability. Providing this service to Some indicators that emergency departments use to a lacker in a short time and giving priority to very evaluate their operations and service quality are GUL, GUNERI Corresponding Author 43 Forecasting ED patient LOS by ANNs average waiting time, average LOS, ED productivity, By the conclusion of the study they point out that resource utilization and layout efficiency [4,5]. ANNs have the potential to be a powerful tool for the Prolonged waiting times have been a major cause of analysis of complex medical system data like EDs. Xu ED overcrowding, that is a main reason of supply– et al. [1] use a data-driven method to identify variables demand mismatch [1]. Prolonged waiting times are correlated with the daily arrivals of the non-critical constituted by the long waits in triage, delays in patients and model this association using ANNs. They testing or obtaining test results, waiting for the compare the ANN modeling with the non linear least physician, and shortage of nursing staff [6]. El-Sharo square regression (NLLSR) and multiple linear [7] reveals several reasons that cause to increase long regression (MLR) in terms of mean average waiting times in emergency departments. These percentage error (MAPE). In this study, we apply a reasons include the inefficient utilization of ED data mining approach ANN to forecast ED patient resources, miscommunication between ED staff to LOS. The constructed model can be used to provide ensure the smoothness of flow of patients in or out of insights to ED medical staff (doctors, nurses etc.) the emergency department, delays in admitting determining patient LOS. The study contributes to the patients from the emergency department, inefficient literature with some aspects. First, it proposes an resource allocation in the hospital, and other external ANN-based forecast model considering new factors, such as an increase in demand of patients due predictors such as mode of arrival, door-to-doctor time to a reduction in the number primary care physicians apart from similar studies in the literature. Second, it in the neighboring area of the emergency department. uses a real data of a leading university hospital ED in LOS is used to assess hospital ED costs and Turkey and the model performed in this ED is the first effectiveness and is made use of several methods to attempt in order to forecast LOS. forecast [8]. Accurate forecasting of patient LOS enables ED management right resource allocation and 3. MATERIAL AND METHOD efficient utilization of the department resources. In [9], it is emphasized that forecasting and determining LOS In this section, the model data, ANNs as the modeling in hospitals can be very useful for hospital technique and constructed ANN model considering management, particularly for prioritizing health care ED LOS forecasts are introduced in sub-sections, policies and promoting health services, comprising the respectively. appropriate allocation of health care resources according to differences in patients’ LOS along with 3.1. Data considering patients’ health status and social- The data is obtained from a regional university demographic features. Therefore, we aim to forecast hospital emergency department in eastern part of patient LOS using Artificial Neural Network (ANN) Turkey that serves approximately an average of within the input factors that are predictive such as 40.000 patients per year. We collect a total data of patient age, sex, mode of arrival, treatment unit, 1500 ED patients who were treated in the department medical tests and inspection in an emergency in October and November, 2010. We provide the department. related data by the aid of hospital information management system, medical staff opinion and 2. LITERATURE REVIEW manual data collecting forms. The variables used for generating ANN model are categorical and numeric. Limited number of studies related to the forecasting of The statistical results for numeric and categorical LOS are available in the literature. The studies use variable types are presented in Tables 1-2. “Number of various methods to forecast including such as ANNs, tests” represents the number of tests that patients take linear regression and logistic regression. Li et al. [8] during their hospitalization in the emergency propose a data mining approach based on Back- department. Most of the patients take one test. “Door- Propagation (BP) neural networks to construct a LOS to-doctor time” represents the time between prediction model. They analyze 921 cholecystitis registration and first treatment area by a doctor. It patients data of a Chinese hospital treated between especially takes 4 minutes. “ED LOS” is our output 2003 and 2007. The model constructed provides variable (dependent variable). It is generally approximately 80% accuracy and reveal 5 LOS considered as the time that passes from the patient’s predictors: days before operation, wound grade, time of arrival at the emergency department until time operation approach, charge type and number of of discharge or transfer to another department of the admissions. Wrenn et al. [10] estimate patient's length hospital. In Table 2, patient mode of arrival express of stay in the ED with an ANN. They develop an the patient type of arrival at the emergency ANN using the software Stuttgart Neural Network department. A patient who comes on foot, by his/her Simulator using patient variables such as age, acuity special car or on a stretcher refers to walk-in patient. level, coded ICD-9 chief complaint, language, order A patient who moves into the ED via an ambulance for a consult service, presence of at least one vehicle (car, helicopter etc.) refers to ambulance laboratory exam, and presence of at least one patient. Treatment areas consist of five different areas radiology exam. They also use operational variables. where patients are sent by their acuity level. GUL, GUNERI 44 Forecasting ED patient LOS by ANNs Table 1. Statistical results of numeric variables. Description Numeric variables affecting Variable ED patient LOS type Detail Min Max Mean SD Number of tests Input Ranges from 0 to 8 0 time 8 times 1,12 times 1,53 Starts with a patient registered and Door-to-doctor time Input 0 min 65 mins 3,96 mins 4,01 ends after seen by a doctor initially Average total length of stay in ED ED LOS Output 5 mins 600 mins 72,15 mins 59,42 per patient Table 2. Statistical results of categorical variables. Categorical variables affecting Variable Number of Detail Occurrence ED patient LOS type records Gender Input (1) Male 51% 770 (0) Female 49% 730 Mode of Arrival Input (1) Walk-in 93% 1396 (0) Ambulance 7% 104 On-call physician Input Practitioner physician (1) 11% 160 Practitioner physician (2) 2% 28 Practitioner physician (3) 13% 189 Practitioner physician (4) 9% 133 Practitioner physician (5) 8% 118 Practitioner physician (6) 11% 158 Practitioner physician (7) 14% 206 Practitioner physician (8) 9% 131 Practitioner physician (9) 11% 167 Practitioner physician (10) 14% 210 Judicial case Input (1) Judicial case 2% 28 (0) Not judicial 98% 1472 Treatment area Input (1) Monitor beds area 53% 799 (2) Emergency-1 25% 375 (3) Emergency-2 17% 262 (4) Emergency response room 4% 57 (5) Resuscitation 0,5% 7 Immediate treatment Input (1) Need immediate treatment 12% 184 (0) Do not need 88% 1316 X-ray Input (1) Taking desired test 22% 329 (0) Not need to take 78% 1171 ECG (Electrocardiogram) Input (1) Taking desired test 16% 238 (0) Not need to take 84% 1262 Hemogram Input (1) Taking desired test 27% 401 (0) Not need to take 73% 1099 Medical biochemistry Input (1) Taking desired test 26% 384 (0) Not need to take 74% 1116 Exact urine analysis Input (1) Taking desired test 2% 34 (0) Not need to take 98% 1466 Tomography Input (1) Taking desired test 6% 84 (0) Not need to take 94% 1416 Arterial blood gas test Input (1) Taking desired test 0,3% 4 (0) Not need to take 100% 1496 Ultrasonography Input (1) Taking desired test 5% 71 (0) Not need to take 95% 1429 Urine test Input (1) Taking desired test 6% 96 (0) Not need to take 94% 1404 Prothrombin time test Input (1) Taking desired test 0,1% 1 (0) Not need to take 100% 1499 Blood center Input (1) Taking desired test 0,1% 2 (0) Not need to take 100% 1498 GUL, GUNERI 45 Forecasting ED patient LOS by ANNs While, critical patients are sent to the beds with This means that the system selects a single hidden monitors (monitor beds area), non-critical patients layer as well as 27 hidden neurons. We make a who have minor injuries, headache, stomach ache etc. correlation analysis before developing an alternative are sent to Emergency-1 and Emergency-2 rooms. ANN model. The alternative model includes factors Patients who had an accident and had major injures that have a strong correlation with the independent are rapidly placed under observation in resuscitation variable LOS. The correlation analysis is performed by area. Patients who have infectious diseases are placed Minitab® 17.1.0 with a significance level (Į ) of 0.05. in a room called emergency response room. We calculate the correlation coefficient for every variable (numeric and categorical) with LOS. The 3.2. Artificial Neural Networks results are given in Table 3. We divide the Artificial Neural Networks (ANNs), which are significance into three correlation strengths as in [8]: generally called as “neural networks” or “neural nets”, strong (1-p>0.95), medium (0.90<1-p<0.95) and weak attempt to reproduce the computational processes (1-p<0.90), where p is the test probability value. taking place in the central nervous system (CNS) by using a set of highly interconnected processing Table 3. Correlation of all the variables. elements [11]. An ANN model, which is formed of n layers, presents a different number of computational Correlation Variables P-value Correlation coefficient elements that function like biological neurons and Gender 0,014 0,583 Weak intensive connections between these computational Mode of Arrival -0,249 0,000 Strong elements among layers. The computational elements On-call physician -0,034 0,185 Weak used in various ANN models are named as artificial Judicial case 0,160 0,000 Strong neurons or process elements [12, 13]. The first layer Treatment area -0,163 0,000 Strong which is called as the “input” layer and the last one Immediate treatment 0,123 0,000 Strong which is called as the “output” layer are used to get X-ray 0,279 0,000 Strong information from inside and outside the network, ECG (Electrocardiogram) 0,239 0,000 Strong respectively. The middle layers which are generally Hemogram 0,539 0,000 Strong called as “hidden” layers are essential to the network Medical biochemistry 0,544 0,000 Strong in order to be able to convert certain input patterns Exact urine analysis 0,138 0,000 Strong into appropriate output patterns [11]. ANNs are Tomography 0,309 0,000 Strong applied to several practices such as forecasting. For Arterial blood gas test 0,089 0,001 Strong ED operations, ANNs are frequently applied to Ultrasonography 0,288 0,000 Strong forecast ED patient arrivals [1], ED LOS [8,10], ED Urine test 0,271 0,000 Strong patient admission process [7], etc [14]. Prothrombin time test 0,014 0,576 Weak Blood center 0,123 0,000 Strong 3.3. Empirical Study Number of tests 0,601 0,000 Strong We prefer to use professional neural network-based Door-to-doctor time -0,005 0,844 Weak software Alyuda Neurolntelligence®. We follow a study process including the steps as here: (1) data 4. RESULTS AND DISCUSSION collection, (2) data preparation, (3) correlation analysis on the variables and (4) modelling by ANN. We use an initial learning rate of 0.1 and a momentum After the data entry to the software, the system of 0.1. We try alternative values for both the learning randomly selects 68% of the data as training sets, 16% rate and momentums of (0.1), (0.2), (0.4) and (0.6) of the data as validation sets and 16% of the data as along with various learning algorithms such as Quick test sets. The software has an automatic architecture Propagation, Quasi-Newton, Online-Back Propagation search module which selects [33-27-1] architecture for and Levenberg-Marquardt. We run all the models training (see Figure 1). within a cycle of 500 iterations. We try 64 various models in total and obtain the best value with the lowest absolute error (Table 4). The scatter plot related to the target, output values of the ANN model and a comparison of actual and output values of the model have been acquired by means of the software (as shown in figure 2). Figure 1. Selection of the best network. GUL, GUNERI 46 Forecasting ED patient LOS by ANNs Table 4. The best network and parameters. Table 5. The best network and parameters (alternative model). Summary of the ANN model Network Architecture [33-16-1] Summary of the ANN model Training Algorithm Levenberg-Marquardt Network Architecture [31-15-1] Hidden FX Logistic Training Algorithm Levenberg-Marquardt Output FX Logistic Hidden FX Logistic Number of Iterations 501 Output FX Logistic Avg. Training Error 25,769552 Number of Iterations 501 Avg. Validation Error 33,430944 Avg. Training Error 25,629 Avg. Test Error 34,287008 Avg. Validation Error 31,319 R-Squared 0,63 Avg. Test Error 34,492 Learning Rate 0,4 R-Squared 0,61 Momentum 0,4 Learning Rate 0,4 Momentum 0,4 5. CONCLUSION This study presents an ED LOS forecasting model using ANNs. We benefit a collected data of 1500 ED patients and identify several variables. While some of them are categorical variables such as gender, mode of arrival, on-call physician, treatment area and taking some tests, the remaining are numerical variables such as number of tests and door-to-doctor time. In the study, we propose two ANN-based models by a base model that includes all variables and an alternative model that takes into consider all the variables except variables with weak or medium correlation on LOS. Each of the models did not give an ideal prediction (a) accuracy that is generally expected to be better than 80% in forecasting. The reason of this is considered to be originated from the selection of insufficient or inaccurate input variables. For future directions, other forecasting techniques such as MLR, logistic regression and support vector machine (SVM) can be used and compared the current results. 6. ACKNOWLEDGEMENT We thank to Dr. Mustafa YÕ ldÕ z, the head of the department of the emergency medicine at FÕ rat University Hospital. And we also owe doctors and nurses work by all three shifts a debt of gratitude for helps on getting access data. “Open Access: This article is distributed under the terms of the Creative Commons Attribution License (b) (CC-BY 4.0) which permits any use, distribution, and Figure 2. (a)Visualization of actual vs. output and (b) reproduction in any medium, provided the original scatter plot of the ANN model. author(s) and the source are credited.” To obtain an alternative ANN model, we exclude the 7. REFERENCES variables that have little affects (variables with weak or medium correlation) on LOS. Therefore we run the [1] Xu, M., Wong, T.C., Chin, K.S., 2013, new model with all the variables except gender, on- Modeling daily patient arrivals at Emergency call physician, prothrombin time test and door-to- Department and quantifying the relative importance of doctor time. The performance of the new model is contributing variables using artificial neural network, shown in Table 5. We could see that the accuracy of Decision Support System, 54, 1488-1498. the models are nearly similar. GUL, GUNERI 47 Forecasting ED patient LOS by ANNs [2] Gul, M., Guneri, A.F., 2012, A computer Comparative Study, International Journal of Industrial simulation model to reduce patient length of stay and Engineering: Theory, Applications and Practice, to improve resource utilization rate in an emergency 15(4), 349-359. department service system, International Journal of [14] Kilmer, R.A., Smith, A.E., Shuman, L.J., 1997, Industrial Engineering: Theory, Applications and An emergency department simulation and a neural Practice, 19(5), 221-231. network metamodel, Journal of the society for health [3] Ersel, M., KarcÕ o÷ lu, Ö., YanturalÕ , S., systems, 5(3), 63-79. Yürüktümen, A., Sever, M., Tunç, M.A., 2006, Emergency Department utilization characteristics and VITAE evaluation for patient visit appropriateness from the patients’ and physicians’ point of view, Turkisj MSc Muhammet GUL Journal of Emergency Medicine, 6(1), 25-35 (In Muhammet Gul is PhD student and has been working Turkish). at the Department of Industrial Engineering, Yildiz Technical University, Istanbul, Turkey. He received [4] Gul, M., Guneri, A.F., 2015, A comprehensive MSc degree in Industrial Engineering from Yildiz review of emergency department simulation Technical University. His research interests are in applications for normal and disaster conditions, simulation modelling, healthcare system management, Computers & Industrial Engineering, 83(5), 327-344. occupational safety, multi-criteria decision making [5] Gül, M., Güneri, A.F., Tozlu, ù ., 2014, and fuzzy sets. Prioritization of emergency department key performance indicators by using fuzzy AHP, 15th Assoc. Prof. Dr. Ali Fuat GUNERI Ali Fuat Guneri has been working at the Department International Symposium on Econometrics, Operations Research and Statistics, Isparta, Turkey. of Industrial Engineering, Yildiz Technical University, Turkey since 1990. He received PhD in [6] Paul, J.A., Lin, L., 2012, Models for Improving Industrial Engineering from the Yildiz Technical Patient Throughput and Waiting at Hospital University. His research interests are in production Emergency Departments, The Journal of Emergency management, supply chain management and Medicine, 43(6), 1119-1126. occupational safety. [7] El-Sharo, M.R.A., 2002, Predicting hospital admissions from emergency department using artificial neural networks and time series analysis, MSc Thesis, Yarmouk University, Jordan. [8] Li, J-S., Tian, Y., Liu, Y-F., Shu, T., Liang, M- H., 2013, Applying a BP neural network model to predict the length of hospital stay, In Health Information Science (pp. 18-29). Springer Berlin Heidelberg. [9] Hachesu, P.R., Ahmadi, M., Alizadeh, S., Sadoughi, F., 2013, Use of data mining techniques to determine and predict length of stay of cardiac patients, Health Informatics Research, 19(2), 121-129. [10] Wrenn, J., Jones, I., Lanaghan, K., Congdon, C.B., Aronsky, D., 2005, Estimating Patient’s Length of Stay in the Emergency Department with an Artificial Neural Network. In AMIA Annual Symposium Proceedings (Vol. 2005, p. 1155). American Medical Informatics Association. [11] Somoza, E., Somoza, J.R., 1993, A neural- network approach to predicting admission decisions in a psychiatric emergency room, Medical Decision Making, 13(4), 273-280. [12] Guneri, A.F., Gumus, A.T., 2008, The usage of artificial neural networks for finite capacity planning, International Journal of Industrial Engineering: Theory, Applications and Practice, 15(1), 16-25. [13] Guneri, A.F., Gumus, A.T., 2009, Artificial Neural Networks for Finite Capacity Scheduling: A GUL, GUNERI http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of Aeronautics and Space Technologies (Havacilik ve Uzay Teknolojileri Dergisi) Springer Journals

Forecasting patient length of stay in an emergency department by artificial neural networks

Loading next page...
 
/lp/springer-journals/forecasting-patient-length-of-stay-in-an-emergency-department-by-CtrBcjseQ8

References (15)

Publisher
Springer Journals
Copyright
Copyright © 2015 by Turkish Air Force Academy
Subject
Engineering; Aerospace Technology and Astronautics
eISSN
2148-1059
DOI
10.7603/s40690-015-0015-7
Publisher site
See Article on Publisher Site

Abstract

10.7603/s40690-015-0015-7 JOURNAL OF AERONAUTICS AND SPACE TECHNOLOGIES JULY 2015 VOLUME 8 NUMBER 2 (43-48) FORECASTING PATIENT LENGTH OF STAY IN AN EMERGENCY DEPARTMENT BY ARTIFICIAL NEURAL NETWORKS Muhammet GUL Ali Fuat GUNERI Yildiz Technical University, Mechanical Yildiz Technical University, Mechanical Engineering Faculty, Department of Industrial Engineering Faculty, Department of Industrial Engineering, Istanbul, TURKEY Engineering, Istanbul, TURKEY mgul@yildiz.edu.tr guneri@yildiz.edu.tr th rd Received: 18 April 2015, Accepted: 23 July 2015 ABSTRACT Emergency departments (EDs) have faced with high patient demand during peak hours in comparison to the other departments of hospitals because of their complexity and uncertainty. Therefore prolonged waiting times in EDs have caused the dissatisfaction on patients. Patient length of stay (LOS), also known as patient throughput time, is generally considered to be the length of time that passes from the patient’s time of arrival at the ED until time of discharge or transfer to another department of the hospital. Starting from patient admissions to the EDs it becomes important have to be known the overall LOS in terms of right resource allocation and efficient utilization of the department. For this purpose this paper aims to forecast patient LOS using Artificial Neural Network (ANN) within the input factors that are predictive such as patient age, sex, mode of arrival, treatment unit, medical tests and inspection in the ED. The method can be used to provide insights to ED medical staff (doctors, nurses etc.) determining patient LOS. Keywords: Forecasting, Patient Length of Stay, Emergency Department, Artificial Neural Networks. YAPAY Sø Nø R Aö LARI KULLANILARAK ACø L SERVø S HASTA KALIù SÜRESø Nø N TAHMø Nø ÖZET Acil servisler hastanelerin di÷ er birimlerine göre karmaú Õ klÕ k ve belirsizli÷ in fazla oldu÷ u birimler olduklarÕ ndan, yo÷ un saatlerde yüksek hasta talebi ile karú Õ laú maktadÕ rlar. Bundan dolayÕ , acil servis içerisinde uzayan bekleme süreleri hastalar üzerinde memnuniyetsizli÷ e yol açmaktadÕ r. Hasta kalÕ ú uzunlu÷ u, di÷ er bir deyiú le hastanÕ n geçirdi÷ i toplam süre, genellikle hastanÕ n acil servise geliú inden taburcu edilmesi ya da hastanenin di÷ er bir birimine sevk edilmesine kadar geçen sürenin uzunlu÷ u olarak de÷ erlendirilmektedir. HastanÕ n acil servise kabulünden baú lanÕ larak, toplam kalÕ ú uzunlu÷ unun bilinmesi do÷ ru kaynak tahsisi ve birimin etkin kullanÕ mÕ açÕ sÕ ndan önemli hale gelmektedir. Bu amaçla bu çalÕ ú ma, hasta yaú Õ , cinsiyet, varÕ ú türü, muayene ünitesi, acil serviste uygulanan tÕ bbi testler ve muayene gibi belirleyici olan girdilerle birlikte Yapay Sinir A÷ larÕ (YSA) kullanÕ larak hasta kalÕ ú uzunlu÷ unu tahmin etmeyi amaçlamaktadÕ r. Metot, acil servis tÕ bbi personeline (doktorlar, hemú ireler vb.) hasta kalÕ ú uzunlu÷ unun belirlenmesi için fikir vermede kullanÕ labilmektedir. Anahtar Kelimeler: Tahmin, Hasta KalÕ ú Süresi, Acil Servis, Yapay Sinir A÷ larÕ . urgent one is essential. But people don't have any 1. INTRODUCTION urgent cure may apply to emergency departments Emergency Departments (EDs) are the busiest rarely. These applications cause patient overcrowding departments in a hospital and its main purpose is to in many emergency departments. Thus waiting times provide timely emergency care to patients [1,2]. The and unsatisfaction levels of patients may increase, a highest attribute of emergency departments is their general complex condition may occur in the ED [3]. uninterrupted serviceability. Providing this service to Some indicators that emergency departments use to a lacker in a short time and giving priority to very evaluate their operations and service quality are GUL, GUNERI Corresponding Author 43 Forecasting ED patient LOS by ANNs average waiting time, average LOS, ED productivity, By the conclusion of the study they point out that resource utilization and layout efficiency [4,5]. ANNs have the potential to be a powerful tool for the Prolonged waiting times have been a major cause of analysis of complex medical system data like EDs. Xu ED overcrowding, that is a main reason of supply– et al. [1] use a data-driven method to identify variables demand mismatch [1]. Prolonged waiting times are correlated with the daily arrivals of the non-critical constituted by the long waits in triage, delays in patients and model this association using ANNs. They testing or obtaining test results, waiting for the compare the ANN modeling with the non linear least physician, and shortage of nursing staff [6]. El-Sharo square regression (NLLSR) and multiple linear [7] reveals several reasons that cause to increase long regression (MLR) in terms of mean average waiting times in emergency departments. These percentage error (MAPE). In this study, we apply a reasons include the inefficient utilization of ED data mining approach ANN to forecast ED patient resources, miscommunication between ED staff to LOS. The constructed model can be used to provide ensure the smoothness of flow of patients in or out of insights to ED medical staff (doctors, nurses etc.) the emergency department, delays in admitting determining patient LOS. The study contributes to the patients from the emergency department, inefficient literature with some aspects. First, it proposes an resource allocation in the hospital, and other external ANN-based forecast model considering new factors, such as an increase in demand of patients due predictors such as mode of arrival, door-to-doctor time to a reduction in the number primary care physicians apart from similar studies in the literature. Second, it in the neighboring area of the emergency department. uses a real data of a leading university hospital ED in LOS is used to assess hospital ED costs and Turkey and the model performed in this ED is the first effectiveness and is made use of several methods to attempt in order to forecast LOS. forecast [8]. Accurate forecasting of patient LOS enables ED management right resource allocation and 3. MATERIAL AND METHOD efficient utilization of the department resources. In [9], it is emphasized that forecasting and determining LOS In this section, the model data, ANNs as the modeling in hospitals can be very useful for hospital technique and constructed ANN model considering management, particularly for prioritizing health care ED LOS forecasts are introduced in sub-sections, policies and promoting health services, comprising the respectively. appropriate allocation of health care resources according to differences in patients’ LOS along with 3.1. Data considering patients’ health status and social- The data is obtained from a regional university demographic features. Therefore, we aim to forecast hospital emergency department in eastern part of patient LOS using Artificial Neural Network (ANN) Turkey that serves approximately an average of within the input factors that are predictive such as 40.000 patients per year. We collect a total data of patient age, sex, mode of arrival, treatment unit, 1500 ED patients who were treated in the department medical tests and inspection in an emergency in October and November, 2010. We provide the department. related data by the aid of hospital information management system, medical staff opinion and 2. LITERATURE REVIEW manual data collecting forms. The variables used for generating ANN model are categorical and numeric. Limited number of studies related to the forecasting of The statistical results for numeric and categorical LOS are available in the literature. The studies use variable types are presented in Tables 1-2. “Number of various methods to forecast including such as ANNs, tests” represents the number of tests that patients take linear regression and logistic regression. Li et al. [8] during their hospitalization in the emergency propose a data mining approach based on Back- department. Most of the patients take one test. “Door- Propagation (BP) neural networks to construct a LOS to-doctor time” represents the time between prediction model. They analyze 921 cholecystitis registration and first treatment area by a doctor. It patients data of a Chinese hospital treated between especially takes 4 minutes. “ED LOS” is our output 2003 and 2007. The model constructed provides variable (dependent variable). It is generally approximately 80% accuracy and reveal 5 LOS considered as the time that passes from the patient’s predictors: days before operation, wound grade, time of arrival at the emergency department until time operation approach, charge type and number of of discharge or transfer to another department of the admissions. Wrenn et al. [10] estimate patient's length hospital. In Table 2, patient mode of arrival express of stay in the ED with an ANN. They develop an the patient type of arrival at the emergency ANN using the software Stuttgart Neural Network department. A patient who comes on foot, by his/her Simulator using patient variables such as age, acuity special car or on a stretcher refers to walk-in patient. level, coded ICD-9 chief complaint, language, order A patient who moves into the ED via an ambulance for a consult service, presence of at least one vehicle (car, helicopter etc.) refers to ambulance laboratory exam, and presence of at least one patient. Treatment areas consist of five different areas radiology exam. They also use operational variables. where patients are sent by their acuity level. GUL, GUNERI 44 Forecasting ED patient LOS by ANNs Table 1. Statistical results of numeric variables. Description Numeric variables affecting Variable ED patient LOS type Detail Min Max Mean SD Number of tests Input Ranges from 0 to 8 0 time 8 times 1,12 times 1,53 Starts with a patient registered and Door-to-doctor time Input 0 min 65 mins 3,96 mins 4,01 ends after seen by a doctor initially Average total length of stay in ED ED LOS Output 5 mins 600 mins 72,15 mins 59,42 per patient Table 2. Statistical results of categorical variables. Categorical variables affecting Variable Number of Detail Occurrence ED patient LOS type records Gender Input (1) Male 51% 770 (0) Female 49% 730 Mode of Arrival Input (1) Walk-in 93% 1396 (0) Ambulance 7% 104 On-call physician Input Practitioner physician (1) 11% 160 Practitioner physician (2) 2% 28 Practitioner physician (3) 13% 189 Practitioner physician (4) 9% 133 Practitioner physician (5) 8% 118 Practitioner physician (6) 11% 158 Practitioner physician (7) 14% 206 Practitioner physician (8) 9% 131 Practitioner physician (9) 11% 167 Practitioner physician (10) 14% 210 Judicial case Input (1) Judicial case 2% 28 (0) Not judicial 98% 1472 Treatment area Input (1) Monitor beds area 53% 799 (2) Emergency-1 25% 375 (3) Emergency-2 17% 262 (4) Emergency response room 4% 57 (5) Resuscitation 0,5% 7 Immediate treatment Input (1) Need immediate treatment 12% 184 (0) Do not need 88% 1316 X-ray Input (1) Taking desired test 22% 329 (0) Not need to take 78% 1171 ECG (Electrocardiogram) Input (1) Taking desired test 16% 238 (0) Not need to take 84% 1262 Hemogram Input (1) Taking desired test 27% 401 (0) Not need to take 73% 1099 Medical biochemistry Input (1) Taking desired test 26% 384 (0) Not need to take 74% 1116 Exact urine analysis Input (1) Taking desired test 2% 34 (0) Not need to take 98% 1466 Tomography Input (1) Taking desired test 6% 84 (0) Not need to take 94% 1416 Arterial blood gas test Input (1) Taking desired test 0,3% 4 (0) Not need to take 100% 1496 Ultrasonography Input (1) Taking desired test 5% 71 (0) Not need to take 95% 1429 Urine test Input (1) Taking desired test 6% 96 (0) Not need to take 94% 1404 Prothrombin time test Input (1) Taking desired test 0,1% 1 (0) Not need to take 100% 1499 Blood center Input (1) Taking desired test 0,1% 2 (0) Not need to take 100% 1498 GUL, GUNERI 45 Forecasting ED patient LOS by ANNs While, critical patients are sent to the beds with This means that the system selects a single hidden monitors (monitor beds area), non-critical patients layer as well as 27 hidden neurons. We make a who have minor injuries, headache, stomach ache etc. correlation analysis before developing an alternative are sent to Emergency-1 and Emergency-2 rooms. ANN model. The alternative model includes factors Patients who had an accident and had major injures that have a strong correlation with the independent are rapidly placed under observation in resuscitation variable LOS. The correlation analysis is performed by area. Patients who have infectious diseases are placed Minitab® 17.1.0 with a significance level (Į ) of 0.05. in a room called emergency response room. We calculate the correlation coefficient for every variable (numeric and categorical) with LOS. The 3.2. Artificial Neural Networks results are given in Table 3. We divide the Artificial Neural Networks (ANNs), which are significance into three correlation strengths as in [8]: generally called as “neural networks” or “neural nets”, strong (1-p>0.95), medium (0.90<1-p<0.95) and weak attempt to reproduce the computational processes (1-p<0.90), where p is the test probability value. taking place in the central nervous system (CNS) by using a set of highly interconnected processing Table 3. Correlation of all the variables. elements [11]. An ANN model, which is formed of n layers, presents a different number of computational Correlation Variables P-value Correlation coefficient elements that function like biological neurons and Gender 0,014 0,583 Weak intensive connections between these computational Mode of Arrival -0,249 0,000 Strong elements among layers. The computational elements On-call physician -0,034 0,185 Weak used in various ANN models are named as artificial Judicial case 0,160 0,000 Strong neurons or process elements [12, 13]. The first layer Treatment area -0,163 0,000 Strong which is called as the “input” layer and the last one Immediate treatment 0,123 0,000 Strong which is called as the “output” layer are used to get X-ray 0,279 0,000 Strong information from inside and outside the network, ECG (Electrocardiogram) 0,239 0,000 Strong respectively. The middle layers which are generally Hemogram 0,539 0,000 Strong called as “hidden” layers are essential to the network Medical biochemistry 0,544 0,000 Strong in order to be able to convert certain input patterns Exact urine analysis 0,138 0,000 Strong into appropriate output patterns [11]. ANNs are Tomography 0,309 0,000 Strong applied to several practices such as forecasting. For Arterial blood gas test 0,089 0,001 Strong ED operations, ANNs are frequently applied to Ultrasonography 0,288 0,000 Strong forecast ED patient arrivals [1], ED LOS [8,10], ED Urine test 0,271 0,000 Strong patient admission process [7], etc [14]. Prothrombin time test 0,014 0,576 Weak Blood center 0,123 0,000 Strong 3.3. Empirical Study Number of tests 0,601 0,000 Strong We prefer to use professional neural network-based Door-to-doctor time -0,005 0,844 Weak software Alyuda Neurolntelligence®. We follow a study process including the steps as here: (1) data 4. RESULTS AND DISCUSSION collection, (2) data preparation, (3) correlation analysis on the variables and (4) modelling by ANN. We use an initial learning rate of 0.1 and a momentum After the data entry to the software, the system of 0.1. We try alternative values for both the learning randomly selects 68% of the data as training sets, 16% rate and momentums of (0.1), (0.2), (0.4) and (0.6) of the data as validation sets and 16% of the data as along with various learning algorithms such as Quick test sets. The software has an automatic architecture Propagation, Quasi-Newton, Online-Back Propagation search module which selects [33-27-1] architecture for and Levenberg-Marquardt. We run all the models training (see Figure 1). within a cycle of 500 iterations. We try 64 various models in total and obtain the best value with the lowest absolute error (Table 4). The scatter plot related to the target, output values of the ANN model and a comparison of actual and output values of the model have been acquired by means of the software (as shown in figure 2). Figure 1. Selection of the best network. GUL, GUNERI 46 Forecasting ED patient LOS by ANNs Table 4. The best network and parameters. Table 5. The best network and parameters (alternative model). Summary of the ANN model Network Architecture [33-16-1] Summary of the ANN model Training Algorithm Levenberg-Marquardt Network Architecture [31-15-1] Hidden FX Logistic Training Algorithm Levenberg-Marquardt Output FX Logistic Hidden FX Logistic Number of Iterations 501 Output FX Logistic Avg. Training Error 25,769552 Number of Iterations 501 Avg. Validation Error 33,430944 Avg. Training Error 25,629 Avg. Test Error 34,287008 Avg. Validation Error 31,319 R-Squared 0,63 Avg. Test Error 34,492 Learning Rate 0,4 R-Squared 0,61 Momentum 0,4 Learning Rate 0,4 Momentum 0,4 5. CONCLUSION This study presents an ED LOS forecasting model using ANNs. We benefit a collected data of 1500 ED patients and identify several variables. While some of them are categorical variables such as gender, mode of arrival, on-call physician, treatment area and taking some tests, the remaining are numerical variables such as number of tests and door-to-doctor time. In the study, we propose two ANN-based models by a base model that includes all variables and an alternative model that takes into consider all the variables except variables with weak or medium correlation on LOS. Each of the models did not give an ideal prediction (a) accuracy that is generally expected to be better than 80% in forecasting. The reason of this is considered to be originated from the selection of insufficient or inaccurate input variables. For future directions, other forecasting techniques such as MLR, logistic regression and support vector machine (SVM) can be used and compared the current results. 6. ACKNOWLEDGEMENT We thank to Dr. Mustafa YÕ ldÕ z, the head of the department of the emergency medicine at FÕ rat University Hospital. And we also owe doctors and nurses work by all three shifts a debt of gratitude for helps on getting access data. “Open Access: This article is distributed under the terms of the Creative Commons Attribution License (b) (CC-BY 4.0) which permits any use, distribution, and Figure 2. (a)Visualization of actual vs. output and (b) reproduction in any medium, provided the original scatter plot of the ANN model. author(s) and the source are credited.” To obtain an alternative ANN model, we exclude the 7. REFERENCES variables that have little affects (variables with weak or medium correlation) on LOS. Therefore we run the [1] Xu, M., Wong, T.C., Chin, K.S., 2013, new model with all the variables except gender, on- Modeling daily patient arrivals at Emergency call physician, prothrombin time test and door-to- Department and quantifying the relative importance of doctor time. The performance of the new model is contributing variables using artificial neural network, shown in Table 5. We could see that the accuracy of Decision Support System, 54, 1488-1498. the models are nearly similar. GUL, GUNERI 47 Forecasting ED patient LOS by ANNs [2] Gul, M., Guneri, A.F., 2012, A computer Comparative Study, International Journal of Industrial simulation model to reduce patient length of stay and Engineering: Theory, Applications and Practice, to improve resource utilization rate in an emergency 15(4), 349-359. department service system, International Journal of [14] Kilmer, R.A., Smith, A.E., Shuman, L.J., 1997, Industrial Engineering: Theory, Applications and An emergency department simulation and a neural Practice, 19(5), 221-231. network metamodel, Journal of the society for health [3] Ersel, M., KarcÕ o÷ lu, Ö., YanturalÕ , S., systems, 5(3), 63-79. Yürüktümen, A., Sever, M., Tunç, M.A., 2006, Emergency Department utilization characteristics and VITAE evaluation for patient visit appropriateness from the patients’ and physicians’ point of view, Turkisj MSc Muhammet GUL Journal of Emergency Medicine, 6(1), 25-35 (In Muhammet Gul is PhD student and has been working Turkish). at the Department of Industrial Engineering, Yildiz Technical University, Istanbul, Turkey. He received [4] Gul, M., Guneri, A.F., 2015, A comprehensive MSc degree in Industrial Engineering from Yildiz review of emergency department simulation Technical University. His research interests are in applications for normal and disaster conditions, simulation modelling, healthcare system management, Computers & Industrial Engineering, 83(5), 327-344. occupational safety, multi-criteria decision making [5] Gül, M., Güneri, A.F., Tozlu, ù ., 2014, and fuzzy sets. Prioritization of emergency department key performance indicators by using fuzzy AHP, 15th Assoc. Prof. Dr. Ali Fuat GUNERI Ali Fuat Guneri has been working at the Department International Symposium on Econometrics, Operations Research and Statistics, Isparta, Turkey. of Industrial Engineering, Yildiz Technical University, Turkey since 1990. He received PhD in [6] Paul, J.A., Lin, L., 2012, Models for Improving Industrial Engineering from the Yildiz Technical Patient Throughput and Waiting at Hospital University. His research interests are in production Emergency Departments, The Journal of Emergency management, supply chain management and Medicine, 43(6), 1119-1126. occupational safety. [7] El-Sharo, M.R.A., 2002, Predicting hospital admissions from emergency department using artificial neural networks and time series analysis, MSc Thesis, Yarmouk University, Jordan. [8] Li, J-S., Tian, Y., Liu, Y-F., Shu, T., Liang, M- H., 2013, Applying a BP neural network model to predict the length of hospital stay, In Health Information Science (pp. 18-29). Springer Berlin Heidelberg. [9] Hachesu, P.R., Ahmadi, M., Alizadeh, S., Sadoughi, F., 2013, Use of data mining techniques to determine and predict length of stay of cardiac patients, Health Informatics Research, 19(2), 121-129. [10] Wrenn, J., Jones, I., Lanaghan, K., Congdon, C.B., Aronsky, D., 2005, Estimating Patient’s Length of Stay in the Emergency Department with an Artificial Neural Network. In AMIA Annual Symposium Proceedings (Vol. 2005, p. 1155). American Medical Informatics Association. [11] Somoza, E., Somoza, J.R., 1993, A neural- network approach to predicting admission decisions in a psychiatric emergency room, Medical Decision Making, 13(4), 273-280. [12] Guneri, A.F., Gumus, A.T., 2008, The usage of artificial neural networks for finite capacity planning, International Journal of Industrial Engineering: Theory, Applications and Practice, 15(1), 16-25. [13] Guneri, A.F., Gumus, A.T., 2009, Artificial Neural Networks for Finite Capacity Scheduling: A GUL, GUNERI

Journal

Journal of Aeronautics and Space Technologies (Havacilik ve Uzay Teknolojileri Dergisi)Springer Journals

Published: Oct 7, 2015

There are no references for this article.